The effect of capital flows on savings versus investment constrained developing countries

The effect of capital flows on savings versus investment

constrained developing countries

Master Thesis

University of Groningen

Faculty of Economics and Business

Name: Koen Harmsen

Student number: 3033228

Email k.harmsen@student.rug.nl

Supervisor: Prof. dr. D.J. Bezemer

Co-assessor: Dr. M.J. Gerritse

1 Abstract

This paper looks at the effect of capital flows on savings versus investment constrained developing countries. The sample used to test this includes data on 79 developing countries over a 25 year period (1990 – 2014). Both fixed-effects models as well as a generalized methods of moments model are employed to test the relation between different types of capital flows and economic growth for these two groups of countries. The results show that there is a positive robust relation between foreign direct investment and portfolio equity inflows with economic growth. Where the effect is larger for savings-constrained economies. For debt flows and capital outflows the results are not robust.

2 1. Introduction

The past decades have seen a surge in globalization and the integration of economies. This led also to an increase in the pressure for countries to open up, but whereas trade liberalization is mostly viewed as benign the same cannot be said for capital account liberalization. Fischer (1998) argued that there are two main arguments for capital account liberalization, the first is that it is an inevitable step in development and the second is that the benefits simply outweigh the costs. This positive view was shared quite widely within academia and policy circles, even international institutions such as the IMF were in favour and made it part of their reform packages. This view also applied to developing countries, opening the capital account was seen as a good strategy towards robust growth. However with time, more and more people began questioning the efficacy of capital account liberalization.

With the passing of time both academia and policy circles began to take a more sceptical approach to the benefits (Benigno, Converse, & Fornaro, 2015). In this view, large capital flows set the stage for financial crises. Furthermore, the impact on economic growth during calm times is also called into question. This increase in doubts about the efficacy of opening up is based for a large part on the lack of positive robust empirical findings. In this paper the focus will be mainly on one critism, given by Rodrik and Subramanian (2009). They argue that opening up to capital flows should only be beneficial to developing countries when they are constrained by a lack of savings. Whereas it could have negative consequences when a country is constrained by lack of investment opportunities. In these countries the negative effects would outweigh the positive ones. Reinhardt (2010) furthermore shows that the complementarity between foreign finance and growth depends on the type of capital flows, with complementarity being the strongest for FDI flows. Other influential authors also call upon more country and context specificity in assessing the merits for a country of capital account openness (Prasad, Rajan, & Subramanian, 2007) (Kose, Prased, Rogoff, & Wei, 2009).

The research question of this paper is:

What is the effect of capital flows on economic growth in savings versus investment constrained developing countries?

In this question we combine the view that different developing countries are held back by different constraints, with the view that different types of flows have different effects. The literature on capital account liberalization is already large. This paper adds to that literature by taking into consideration that not all poor countries are held back by the same factors (which is assumed in most models). This also means that this paper is relevant for policymakers and the discussion whether a country should open up. If a country is savings-constrained then opening the capital account could have an overall positive effect, but the reverse could be true for investment constraint countries. An example of such a negative effect is in the appreciation of the currency, as a result of capital inflows. This hurts the economy as explained by Rodrik (2008). Moreover, the question of opening up is also linked to the financial trilemma. Rey (2015) argues that this trilemma is actually a dilemma,

3 means that with capital account liberalization a country also gives up its independent

monetary policy. These points suggest that the consequences of liberalization can be far reaching. It is therefore imperative for developing countries to know under what

circumstances an open capital account can help in their development. In answering the research question, we focus on the following hypothesis:

Capital inflows have a more positive effect on economic growth in savings constrained economies compared to investment constrained economies.

To differentiate between saving and investment constraint economies, we look at the

correlation between total gross capital inflows and investment in a country. After having split the sample, we use data over 25 years for 79 developing countries in a fixed effect panel model to test the relationship with economic growth. We furthermore use a generalized methods of moments estimation to check the robustness of the results. The results partially confirm the hypothesis, foreign direct investment inflows and portfolio equity inflows have a positive effect on economic growth. Where the coefficients are higher for savings-constrained economies. However in both types of economies, portfolio debt flows and other investment flows are not significant, and sometimes have a negative impact on economic growth.

Therefore, the conclusion is that the impact of capital flows is quite similar between savings- and investment-constrained economies. With a positive effect of equity flows and no robust relation for debt flows.

4 2. Literature review

In this section we will place the research question within the relevant literature. The literature on capital account liberalization, capital flows and its effect on economies is vast. We will first look at the theoretical case for capital account liberalization, then look at alternative theoretical views before going to the empirics. In the last part we will look at some literature concerning our methodology.

The dominant paradigm, which is neoclassical, on long run economic growth states that the central limiting factor to economic development is the lack of capital endowment in less developed countries (Dullien, 2009). Both growth models based on Solow and newer endogenous growth models reach this conclusion. In these models output is a function of production factors capital (K) and labor (L). New growth models focus more on human capital. However, investment in human capital is then still constrained by the available resources for investment, as capital in these models can only be increased by investment. Within a country, savings are a necessary condition for investment. If people refrain from consumption the amount of funds available for investment within the economy increases. A balance between demand and supply of funds is reached by the interest rate, so more demand leads to higher interest rate which leads to more savings, and vice versa.

In these models, there are only two ways to increase investment (achieve a higher capital stock). The first is more savings, which are available when people refrain from consumption. The second way is to import savings from abroad, thus capital inflows. The prediction then is that capital inflows should help economic growth as capital moves to countries with better investment opportunities, and as a source of technological spill-overs (Gente, León-Ledesma, & Nourry, 2015). The direction of capital flows, should be from countries with low to

countries with high autarky returns. These returns depend on capital scarcity and long run growth prospects, which are exogenous and country specific (Gourinchas & Rey, 2014). Combining these elements, the theory points out that capital should flow from advanced economies with lower returns toward emerging economies with higher returns and higher growth prospects (Reinhardt, Rici, & Tressel, 2013).

Reinhardt (2010) states this result in terms of productivity, in the neoclassical growth model countries with stronger productivity growth should attract more capital inflows, as strong productivity growth increases the marginal product of capital which makes investments more profitable. In the extreme (under the assumptions that all countries use the same constant returns to scale technology), if capital were allowed to flow completely free, new investments would occur only in poorer countries. This would only stop once the return on investment was the same in all countries (Alfaro, Kalemli-Ozcan, & Volosovych, 2008). On a global level, free capital movement would facilitate an efficient global allocation of savings and

investment in productive uses (Fischer, 1998). This in turn would lead to optimal economic growth and welfare.

5 diversification. A last benefit is that financial flows could foster financial sector development by imposing discipline (Kose, Prased, Rogoff, & Wei, 2009).

One of the assumption behind the reasoning above is that poor countries are poor due to lack of investments, and therefore have high autarky returns. Rodrik and Subramanian (2009) argue in their paper that there are lots of things ‘wrong’ in poor countries and that trying to fix everything at once is impossible. This implies that policies should focus on the correct

constraints for that economy at that point. A way of looking at the binding constraint is by using a decision tree (Hausmann, Rodrik, & Velasco, 2007). The top question/decision in this tree is: “is private investment in the economy held back primarily by lack of access to finance or by low perceived returns?” (Rodrik & Subramanian, 2009). The first constraint means that a country is savings constrained, there is low investment but there are many profitable private projects that are not financed because entrepreneurs cannot get credit. The second type of economy is said to be investment constrained, there investment is low despite there being enough credit, the constraint here is that entrepreneurs do not see many profitable investment opportunities. But if a country is savings constrained, then foreign capital inflows should be beneficial for growth.

Whereas Rodrik (2009) argues that capital flows are not beneficial for all developing

6 Another assumption of the neoclassical model, which is often questioned, is that foreign funds can directly be used in the recipient country. And that after opening up foreign investors will invest in the country. However, certain thresholds have to be met in a country before investors are willing to invest and before foreign funds can be efficiently used. These thresholds are closely linked to institutions, which have gained more and more importance in the

development literature (Acemoglu & Johnson, 2005). Alfaro et al (2008) have found that indeed institutional quality has shaped international capital flows over the last decades, they argue this could be one of the links through which institutions affect long run development. Two points about thresholds are important, the first is that certain reforms have to be taken before foreign investors are willing to invest in the country, this means that liberalization may become more successful if combined with other reform measures such as those concerning regulation or financial markets (Kose, Prasad, & Wei, 2010). The second point has to do with the effect that capital inflows can have on reform processes, foreign investors can add

discipline to those within a country. Those two effects together suggests that international integration or opening up might have different effects on countries with different institutions or institutional development (Braun & Raddatz, 2007). Bumann et al (2013) find that

empirical research on the effects on growth of financial liberalizations combined with other policies is hardly available.

Next we will look at some specific thresholds identified within the literature. The first has to do with legal protection and property rights. Foreign investors will not invest in poorer

countries, even if there are profitable investment opportunities, if contracts cannot be enforced or they face a high risk of expropriation (Reinhardt, Rici, & Tressel, 2013). The second

threshold is financial (system) development, Prasad et al (2007) argue that with a less developed financial system the absorptive capacity for capital inflows is low. A similar argument is made by Ahmed and Zlate (2014), large capital inflows according to them can overwhelm the intermediation capacity of the domestic financial system. This can lead to excessive credit creation and asset price bubbles, which are bad for economic growth and could lead to financial instability. However, capital account liberalization could also have a positive effect on financial development and subsequently economic growth. Braun and Raddatz (2007) make the point that not only do capital inflows help to develop domestic financial markets and institutions, it also provides firms with easier excess to international financial markets for funds. Domestic firms can then leapfrog the domestic financial system, this is obviously not possible for all firms. Low domestic financial development could

constrain domestic demand (too high precautionary savings) as residents are unable to borrow against future income or store value in sound financial instruments (Gourinchas & Jeanne, 2013). If capital flows increase the efficiency of the financial system, by reducing information asymmetries / pooling risks, then these constraints could be resolved. A more efficient

7 corrupt countries, capital account liberalization will have a negative effect on growth

compared to less corrupt countries as in more corrupt countries the benefits will be usurped. So far we saw have seen two different theoretical perspectives and the effect of institutions (threshold effects). Whereas the neoclassical theory was very positive on capital account liberalization, the Schumpeterian-Keynesian focuses more on the negative side as they do not see capital inflows as necessary to achieve higher levels of investment. Even in this more negative view however, capital inflows could still have a positive effect on growth and development. Especially in countries that are savings constrained. Not all developing countries will be able to start the process of credit creation as trust/stability is missing or certain institutions are underdeveloped.

We could therefore say that capital account liberalization, for all theoretical perspectives, could help solve problems with financing constraints. For example, equity market

liberalization (part of capital account liberalization) could reduce financing constraints by making foreign capital more readably available (Bekaert, Harvey, & Lundblad, 2005). Indirectly, institutions could also improve, as foreign investor would bring discipline and a need for better governance. Such improvements would mean that both internal and external finance would be reduced in costs. In this indirect way, higher financial development through opening up would could lead to economic growth.

The last part of the theory on capital account liberalization focuses the negative effects of opening up. The first of these, again is from Rodrik (2008). In this paper he argues that there is a strong relationship between the exchange rate and economic growth, simply put

undervaluation stimulates growth whereas overvaluation hampers growth. He estimates this to be a quite linear relationship, an increase in undervaluation has the same effects as a decrease in overvaluation, this relationship only holds for developing countries. This view counters that developing countries are constrained by finance and that capital flows have a positive effect. An increase in capital flows could lead (especially if it are large inflows) to an appreciation of the currency (Ahmed & Zlate, 2014) (Borio & Disyatat, 2011). This would hurt exports and economic growth performance. A similar argument is also made by Prasad et al (2007), although they focus only on the negative effect of appreciation and not the positive of undervaluation. Another negative aspect that capital account liberalization brings is that too much or the wrong kind of inflows (debt) could lead to imbalances, we saw that Borio and Disyatat (2011) made this argument of excess elasticity and a similar argument is put forward by Rey (2015) as the global financial cycle. If a country opens up, it will lose control and be dependent on global forces that shape global liquidity and international capital flows. 2.2 Empirics

Kose et al (2009), give an overview of the empirical literature on the causal relationship between openness and growth. They find mixed results, some authors do find a positive relationship whereas others find nothing or a negative relationship, especially after controlling for usual determinants of economic growth. In this section we will first look at the direction of flows as the neoclassical model is very clear on this. After which the focus will shift on

8 As we have seen, standard international macroeconomics predicts that capital should flow from (capital) rich to poor countries where the marginal productivity of capital is higher (von-Hagen & Zhang, 2014). Also, there should be no (or little) difference between gross and net flows. However, since the 1990’s when Lucas observed that too little capital flows from rich to poor countries, the validity of this assumption underlying the neoclassical model have been questioned (Alfaro, Kalemli-Ozcan, & Volosovych, 2008). This result, that capital flows are not in the direction predicted by the neoclassical model is found again and again in the literature, such as in Prasad et al (2007) and Reinhardt (2010). Gourinchas and Jeanne (2013) find that capital flows from rich to poor countries is not only low in volume, but their

allocation across developing countries is uncorrelated with the predictions of standard textbook model. This negative correlation between growth and capital flows could be due to the differential effect of public and private flows (Gourinchas & Rey, 2014), whereas in contrast to public flows private flows are positively related with productivity variables. Von Hagen and Zhang (2014) give as reason for the uphill net capital flows that rich countries ‘export’ their superior financial services to developing countries.

Reinhardt et al (2013) however find that amongst countries with an open capital account, richer countries experience net capital outflows and poorer countries net inflows. They also find that capital account restrictions are effective, among countries without capital account liberalization there is no systemic relation. The results of Reinhardt et al (2013) suggest that development finance was an important driver of international capital flows amongst open countries. To assess the extent of development finance, Aizenman et al (2007) looks at the self-financing ratios of developing countries and do not find significant difference before compared to after liberalization. They argue that this is consistent with the notion that financial integration leads to greater diversification of assets and liabilities, with inflows of foreign savings and outflows of domestic savings. However, earlier research, such as Obstfeld (1995) and Reinhart and Talvi (1998) find that net foreign resource inflows are negatively related to national saving and positively related to domestic investment.

The most important empirics for this paper have less to do with the direction of flows but more what the effects of capital flows are on economic growth. The mixed results in these studies are partially due to the different possibilities of measuring capital flows. For example, different types of liberalizations could have different effects, capital account liberalization is opening up the economy to all types of flows, whereas equity market liberalization limits this to FDI/equity flows. A further complication is that some studies look at financial

liberalization, which also can include liberalization of the domestic financial system (Brafu-Insaidoo & Biepke, 2014).

9 of a financial development threshold. The most positive finding on the effects of capital account liberalization is the study by Bekeart et al (2005). They focus on equity market liberalization and find that after liberalization average annual economic growth in these countries is 1% higher. And this is after controlling for other variables that determine economic growth. Their results are also in line with the neoclassical model, in that after liberalization risk sharing increased, cost of capital decreased and investment increased. However, the authors also mention that their results are surprising given alternative theories and empirical studies that fail to find a robust relation. Laeven (2003) looks at the firm level in 13 developing countries to find out whether liberalization relaxes financing constraints for firms. His results, using panel data, suggest that the liberalization process has a different effect on small versus large firms. Small firms, that before linearization couldn’t get funds from abroad, were more constraint and became less so after liberalization. For large firms, however, constraints became higher after liberalization, the explanation given by Laeven is that these large firms lost access to preferential cheap credit that they enjoyed before liberalization.

Quinn and Toyoda (2008) use a large sample of both developed and developing countries to test the relation between a de jure measure of capital account liberalization and economic growth. They find that capital account liberalization has a positive effect on growth (in the period studied) for both developed and emerging economies. They also find that equity

market liberalization has an independent effect on growth, in line with the result from Bekeart et al (2005). These last results are also in line with an earlier study by Gruben and Mcleod (1998), who find that for a sample of 18 developing economies an increase in both FDI flows to GDP and equity flows to GDP have a significant and positive effect on growth. The

positive effect of FDI is also found by Harrison et al constraints (2004), who look at the effect of capital flows on financing constraints on firm level, and find that FDI is associated with a lowering in financing constraints. Furthermore, they show that restrictions on the capital account effect financing constraints negatively. Alfaro et al (2014) furthermore find that net international private capital flows are positively correlated with productivity growth. Probably the most influential study that does not find a positive relationship is Prasad et al (2007). They find no evidence that an increase in foreign capital inflows directly boosts growth. They argue that these results are due, and provide some evidence for, that the absorptive capacity for foreign resources is low. This could be due to underdeveloped

financial markets or because of overvaluation, two concerns that where already raised earlier. The findings of Prasad et al (2007), suggests that foreign capital inflows do not hurt growth in poor countries but they do not help either. A similar conclusion is reached by Gente et al (2015), they fail to find a robust association between flows and growth. They argue that countries are constrained by investment not savings, and foreign capital cannot be used very well. This is supported by Aizenman et al (2007) who find that greater financial integration in international markets has not led to a situation in which foreign savings, on average, have proved to be a viable source of financing domestic capital for developing countries. Kunieda et al (2014), study the relation between openness and growth and the interaction with

10 predictions, namely that, their interaction term between financial openness and corruption has a negative (and significant) effect of economic growth. Their interpretation of these results is that financial openness magnifies the negative effects of corruption on economic growth. One of the more important results that empirical studies find, when they make the distinction, is that different types of flows have different effects. Henry (2007) argues that at least a distinction has to be made between debt and equity flows, he sees short term debt flows as causing trouble whereas for equity flows recipient countries derive substantial benefits. This view seems to empirically robust, whereas the growth benefits of broad capital account liberalization are not robust, authors that use de facto flows do find results. For example Aizenman et al (2013) find that the relationship between growth and capital flows depend on the type of flow, with large and robust relation with FDI, a smaller and less stable one for equity flows and no relation or a negative one (dependent on the period under study) for debt flows. Dhingra, (2004) using a dataset of 58 developing countries finds the same results. Equity flows, which include FDI and portfolio equity flows do have a positive effect on output growth compared to debt flows which are more volatile and have no effect. A similar result is found by Reisen and Soto (2001).

Reinhardt (2010) furthermore shows, with a cross sectional analysis of countries, that the correlation between capital inflows and productivity growth is positive only for FDI, and negative for equity and debt flows.

2.3 Literature on methodology

Next the focus shifts to the literature concerning two parts of our methodology. The first has to do with the distinction between savings and investment constrained economies and how to makes this distinction. The second issue has already been touched upon, and has to do with what measurement to take for capital flows.

Rodrik and Subramanian (2009) use the US interest rates to elucidate whether certain countries are savings or investment constrained. To check this, they look at the correlation between domestic investment and the US interest rate (their proxy for capital flows). A country is then savings constrained if this relation is negative, a higher US interest rate would lead to lower domestic investment. Banerjee et al (2016) also show that a US contractionary monetary policy shock leads to a retrenchment in capital flows to emerging markets and a subsequent fall in GDP. However, over the last couple of years, other ‘proxies’ for capital flows have been introduced, the VIX being the most important one.

11 countries’ specific macroeconomic conditions. Bruno and Shin (2015) make a similar

argument, when global banks apply more lenient conditions on local banks, the more lenient credit conditions are transmitted to the recipient economy. In this way, more permissive liquidity conditions in the sense of greater availability of credit will be transmitted across borders through the interactions of global and local banks.

Calvo et al. (1996) made the distinction between the global “push” factors for capital flows from the country-specific “pull” factors, and emphasized the importance of external push factors in explaining capital flows to emerging economies in the 1990s. Forbes and Warnock (2012) also find that global factors, especially global risk are associated with extreme capital flow episodes. And that domestic characteristics are generally less important.

Another important part of the empirical literature focuses on how to measure capital account liberalization. We follow Kose et al (2009) who argue that the distinction between de jure and de facto integration is very important. Countries can have capital controls on paper that are not used in practice, therefore it is better to use de facto integration as measured by capital flows or stocks of foreign assets and liabilities. We will use a de facto measure, namely capital flows. For capital flows, a distinction can be made between gross and net flows. Forbes and Warnock (2012) argue that analysis based on net flows would have been fine a few decades ago but that now they would miss the dramatic changes in gross flows and ignore the information in these flows. Flows form domestic investor have become more important, and changes in net flows can no longer be interpreted as being driven by only foreigners. Furthermore, according to Borio and Disyatat (2011) the current accounts tell us little about the role a country plays in international borrowing, lending, intermediation, and the degree to which investment in financed from abroad or domestic financial conditions. For example, if a country’s current account is in balance, the whole of its investment expenditures may be financed from abroad, by loans for example. A current account balance only implies that domestic production equals domestic spending, not that domestic saving finances domestic investment. They also argue that developments in gross capital flows during the financial crisis confirm that net capital flows do not capture the severe disruption in cross-border interbank lending nor do they correctly predict the source of strains (Borio & Disyatat, 2011) A last distinction that can be made for capital flows, and which is very important for this paper is the decomposition of flows. From the literature we know that different flows can have different effects on economic growth. Aizenman et al (2013) e.g. finds positive effects for FDI (robust) and equity flows (less stable) and no effect or a negative relation for debt flows. Dullien (2009) notes that this is not so strange, as some of the big benefits of

12 2.4 Main arguments and the causality of different types of flows

In the literature review many arguments on the effects of capital flows were made, in the table 1 we lists the main positive and negative effects of capital flows. We also show the thresholds that were mentioned in the literature. Some of the effects overlap with each other such as number 2 and 6 with the positive effects.

Positive Negative Thresholds

1. Technological spill overs

1. Credit build-up 1 Institutional quality 2. Better resource

allocation

2 Asset price bubbles 2. Corruption 3 Portfolio diversification 3. Appreciation exchange rate 3 Financial development 4. Institutional development (indirect effect) 4 Financial imbalances 5 Lifting financing constraints (adding to domestic savings)

5. Sudden stop capital flows (abrupt

reversibility) leads to bankruptcies,

destroys local credit channels 6. Increase in efficiency (allocation, competition, deepening domestic market, reducing capital costs. 6 Distorted consumption patterns. 7 Improved risk sharing

Table 1 : Summary of the effect of capital inflows on economic growth

These are however more general statements and we have already seen that different types of flows can have a different impact. Or as Kose (2010) puts it, not all types of capital flows are created equally. Therefore, we will summarize the way through which (the causal relation) certain inflows could affect economic growth at all. Turner (2016) makes the point that an important distinction to make is that between equity flows (FDI and portfolio equity) and debt flows (portfolio debt and other investment flows). The main reason for this distinction lies in the reversibility of these flows. Equity flows are stable compared to debt flows, which are more influence by global and local uncertainties.

13 Portfolio equity flows are seen after FDI as the most beneficial type of flow. Kose et al

(2010), find mainly positive effects on economic growth from this type of flow, although they argue that this could be partially due to simultaneous other reforms when opening up.

14 3. Methodology

To answer the research question (and the hypothesis) we take two steps in the methodology. The first step of the methodology is focused on classifying countries as either savings- or investment-constrained. Rodrik and Subramanian (2009) do this by looking at the correlation between domestic investment and the US interest rate as we saw before. They argue that if all poor countries are poor due to lack of investment and opening up would be good for them then you would expect a negative correlation between these two variables. As this would mean that if US interest rates increases (money flows out of developing countries towards the US) domestic investment would decline as funds disappear. On the other hand, if US interest rate declines then more money possible flows towards developing countries and domestic investment would increase. They give a few example and for the countries they choose only China and India seem to be savings-constrained.

In the literature review we also saw that in the last couple of years the VIX is more and more used by researchers who look at capital flows / global financial cycle. The VIX is an index that is used as a proxy for global uncertainty / global liquidity. Where higher values of VIX mean more uncertainty and therefore less global liquidity.

Both measure, however, are proxies for capital inflows. In this paper we will compare the correlation of the VIX/US interest rate with domestic investment, but also take the correlation between total gross inflows of foreigners divided by GDP in the recipient country with

domestic investment. The investment figures are taken from the World Bank and measure gross capital formation as a percentage of GDP, see Appendix B for a full definition. The last measure, correlation between total gross capital inflows and investment, is our preferred measure as it takes into account not just global factors (which are important for capital flows) but also domestic characteristics. This will therefore give us the best idea if a country is held back by lack of investment opportunities or by a lack of funds. For the VIX,

savings-constrained countries would have a negative correlation with investment, whereas for the capital inflow measure this would be a positive correlation with investment. In appendix A, we show the table for all countries in our sample with the results from the three different measures.

15 In this first step of our methodology, a cutoff-point of 0.5 is taken. Which means that

countries where the correlation between gross capital inflows over GDP and investment is >0.5, they are classified as savings-constrained. And <0.5 as investment-constrained. The graph below shows the distribution of the correlations. Different possible ways to classify the binding constraint are possible, such as a continuous measure. However, splitting the sample in two with a dummy for savings-constrained economies has also some advantages. It allows for better comparing the two groups with descriptive statistics, and in the regressions. A continuous measure could make a distinction between mildly savings-constrained economies and very savings-constrained economies but it is not clear that this would be linear. That a step from a correlation of 0.1 to 0.2 should have the same impact as one from 0.6 to 0.7.

Figure 1: Correlation between gross capital inflows over GDP and domestic investment

The second step after splitting up the sample is to run the model which is shown below. The dependent variable in this model is economic growth (Y= real GDP per capita), with X being control variables and 𝛽₃being the coefficient of interest. We look at the different gross flows over GDP to test what their effect is on economic growth. Both gross capital inflows and outflows are included.

The interaction term after the gross flows over GDP, can stand for either a dummy for

savings-constrained economies (D) or for possible interaction with other variables. In our case these could be the thresholds identified in the literature. 𝛽₁ is for the lagged dependent, we include this because countries that have grown faster in the recent past are more likely to grow faster now, for example the very rapid economic growth of China over the past decades.

∆𝑌_𝑖𝑡 𝑌𝑖𝑡−1 = 𝛼𝑖𝑡 + 𝛽0+ 𝛽1 ∆𝑌_𝑖𝑡−1 𝑌𝑖𝑡−2 + 𝛽2𝑋𝑖𝑡 + 𝛽3( 𝐹𝑙𝑜𝑤 𝐺𝐷𝑃 ∗ 𝐷/𝐼)𝑖𝑡−1 + ∆𝜖𝑖𝑡 X stands for macroeconomic control variables, which are listed in the data section. One of these controls is initial GDP per capita, is included because countries that are richer grow slower due to less opportunities to catch up.

In the literature there is no consensus about whether to take the controls variables as levels or as changes (the equation shows levels). Therefore we will test the baseline model, with only controls, for both cases and see what the best fit is. However, beforehand levels are preferred because the e.g. the level of corruption probably has more influence on economic growth than the change in corruption.

16 We use lagged gross flows over GDP in our model, the papers in our empirical literature review use a lag of one period or one year. Although there is consensus that a lag should be used, there is not clear reason why this should be one year for example instead of two. We, therefore, also test with different lag structures to see what fits best.

The basis specification can be tested in multiple ways. The first is as an interaction model, where the different flows are included individually and with as an interaction with the constrained dummy. Secondly, we can run the regressions separately for the savings-constrained group and the investment-savings-constrained group. All these regressions will make use of a fixed effects model.

A common approach within the literature is to average periods, instead of just using the annual data. This approach is taken because it smooths e.g. the effects of business cycles, and limits the impact of one off events. We look at averaged periods of 4 and 6 years, which are non-overlapping. The data also allows for splitting up the sample in many different ways to test the effects for certain groups or periods. These kinds of ‘extra’ analysis will be performed as robustness checks.

Another robustness check is a general methods of moments estimation. A number of papers looking at the effects of capital flows use this technique, one of the advantages of it is that it can deal with endogeneity problems. That is, do capital flows lead to economic growth or does economic growth lead to capital flows. In our basic specification we try to deal with this by taking lagged capital flows. The Arellano-Bond estimation (GMM) however allows us to use instrumental variables for this. This procedure was first introduced by Arellano and Bond (1991), and is one of the preferred specifications in the literature on capital account

17 4. Data

In this section we will describe the dataset that is used in the model specified in the last section. We will first look at the countries and years that make up the sample, after which we will go through the variables that are used. The description of these variables include how they are measured and what the expected sign is. For a full description with the relevant sources, see Appendix B. After describing the variables, we will look at the descriptive statistics.

The sample of countries is mostly based on data availability. Because the focus is on

developing countries, a lot of countries fell out of the sample due to missing data on some of the variables used. In the end we are left with a sample of 79 countries. See appendix A for the list of countries. 23 are classified as savings-constrained and 56 as

investment-constrained.

Growth of real GDP per capita

This is our dependent variable, and the data to construct it come from the PENN world table. We take real GDP at constant 2011 national prices , and divide by population figures which are also from the PENN world table. This gives us real GDP per capita, of which we then take the growth rate. About half of the empirical studies from the literature review use real GDP per capita growth, the other half use real GDP per capita corrected for PPPs. Our measure thus takes into account inflation but not the relative price levels between countries.

Initial GDP per capita

We include initial GDP per capita, because countries that are richer have less opportunities for catch-up growth and are therefore more likely to have a lower growth rate. The data is the same as for the growth rate but here we take the natural logarithm of GDP per capita.

Exchange rate stability index

This index shows a value between 0 and 1, where higher values of the index indicate a more stable movement of the exchange rate against the currency of the base country. The index is created by Aizenman et al (2010). They measure the stability by looking at the annual standard deviations of the monthly exchange rate between the home and the base country. They also apply a certain threshold to these movements. The expectation is that a more stable exchange rate is good for economic growth.

Monetary independence index

The monetary independence index is also set up by the same authors as the exchange rate stability index. Higher values here, indicate more monetary policy independence. They measure this index as the reciprocal of the annual correlation between the monthly interest rates of the home country and the base country. Again more monetary policy independence should be beneficial for economic growth.

Savings

18 Bank. The expectation is that there is a positive relation between savings and economic

growth.

Investment

The data on investment is also taken from the World Bank. Here again, more investment should be beneficial for economic growth. However, because we are interested in the effect of capital flows on economic growth, and we have seen that certain capital flows work mostly through investment, we will often not include this variable in the model.

Corruption

For corruption we use the Bayesian corruption indicator form the Quality of Government Institute (2017). The value of this index lies between zero and hundred, with higher values meaning more corruption. The value of this index will change every year as new information becomes available. As a higher value indicates more corruption, we expect a negative relation here with economic growth.

Quality of government (institutional quality)

This is the main indicator from the quality of government institute, which we will take a proxy for institutional quality. This index is between 0 and 1, where higher values indicate higher quality of government, the value is the mean of variables “Corruption”, “Law and Order” and “Bureaucracy”. The main reason for using this composite index has to do with data availability, this index is available for most countries in our sample where other possible variables were not. The relationship with economic growth is a bit ambiguous. Institutional quality will be beneficial for economic growth till a certain level. However, after this ‘threshold’ is reached it is not obvious that more quality will lead to more growth.

Furthermore, high institutional quality is often found in developed countries where economic growth is lower.

Trade openness

As mentioned in the introduction, most scholar and international organizations do agree over the benefits of opening up a country to trade, whereas for capital account liberalization there is no consensus. To measure trade openness we take the variable merchandise trade as a share of GDP from the World Bank. The relation between trade openness and economic growth is expected to be a positive one.

Credit

Credit, which is private credit by deposit money banks and other financial institutions as a percentage of GDP comes from the World Bank. This variable is used in two ways, the first is just a credit, where one would expect that more credit, till a certain point would be beneficial for economic development and growth. However, recent studies such as Kunieda et al (2014) find a negative relation between credit and economic growth The second way is using credit as a proxy for financial development, this is important as we saw in the literature because financial development is one of the thresholds that is often mentioned..

Human capital index

19 schooling and returns to education. The expected relation with economic growth is a positive one, higher educated people are more likely to productive.

Capital flow data

In this paper we look at gross capital flows and we follow the decomposition of flows from Broner et al (2012). This decomposition is shown in figure 2. Total gross capital flows consist of capital inflows by foreign agents (CIF) and capital outflows by domestic agents (COD).

Figure 2: Decomposition of capital flows

A further distinction can be made between the different types of flows, we look separately at foreign direct investment, portfolio investment (equity and debt) and other investments (bank flows, loans and trade credit). Where liabilities are inflows from foreigners and assets

outflows by domestic agents.

The authors at the World Bank set up a database covering these different types of flows, but this database did not have the correct countries nor periods, therefore, we use data from the IMF balance of payments statistics. However, we do follow the methodology used by the authors.

All flows are divided by GDP in our analyses. Next to the flows themselves, the variable capital account openness is also ‘made’ from this data. Capital account openness is total gross flows divided by GDP. The inflows for the interactions with the thresholds are total gross capital inflows by foreigners divided by GDP.

Gross flows (total)

CIF

FDI liabilities _liabilitiesPortfoio

Portfolio

equity Portfolio debt

Other investment

liabilities

COD

FDI assets Portfolio _assets

Portfolio

equity Portfolio debt

Other investment

20 4.2 Descriptive statistics

Now we will look at the descriptive statistics, table 2 summarizes the variables used. The sample is split here into investment-constrained and savings-constrained to make comparison easier.

Table 2: Summary statistics

From this table we can see that in many respects the two groups are similar. Real GDP per capita is quite close, with the growth rate in the savings-constrained group being higher. Interestingly, the savings-constrained part has a higher savings rate in the period considered, and a bit higher investment rate as well.

This table also shows quite a few possible outliers. The minimum and maximum observations for economic growth are very low and very high. The extreme negative economic growth is due to some African and Eastern European countries in the beginning of the 1990s. The positive outliers are again a few African countries, but now in the 2000s. This could be due to changes in models used by the statistical agencies of those countries, were revision led to a jump in GDP (Coyle, 2014).

For the capital flow variable there also seem to be outliers in the set, especially in the savings-constrained part. The minimum and maximum values for FDI inflow, FDI outflow, PE inflow and PE outflow are very large. Upon closer examination all these outliers are due to one country: Mauritius. We will therefore not include this country in the analysis.

name obs mean Std. Dev. Min Max obs mean Std. Dev. Min Max

Real GDP per capita 1400 7.876 6573,788 412,0864 34539,55 575 6631,256 4573,128 779,7772 20314,73

Log real GDP per capita 1400 8,549111 1,009529 6,021233 10,44986 575 8,514416 0,8099942 6,659008 9,919102

Economic growth 1344 0,020937 0,0522912 -0,46214 0,354022 552 0,02809 0,0516616 -0,24693 0,329334 ERSI 1374 0,531453 0,3115901 0,007464 1 543 0,60963 0,3348594 0,014229 1 MII 1299 0,479982 0,1741596 0 0,958325 523 0,489857 0,1636662 0,010561 0,944869 Savings (%of GDP) 1333 18,75297 10,30847 -36,6624 77,34215 526 21,00683 11,62738 -24,0038 57,47493 Investment (%of GDP) 1397 22,30868 7 0 54,48614 575 23,97658 9,332011 -2,42436 58,15072 Corruption 1344 52,29246 9,037711 28,43715 69,00919 551 54,84591 4,322369 44,3711 65,58304 QoG 1225 0,474575 0,1452779 0,060185 0,944444 508 0,462029 0,108548 0,12963 0,736111

Trade (Exports + imports /GDP) 1374 59,09579 29,65714 7,994425 192,1234 565 54,35448 30,94199 9,051853 160,0745

Human capital index 1325 2,206068 0,6570462 1,056958 3,713929 500 2,044052 0,5165094 1,029605 3,128288

Credit (%of GDP) 1345 31,60294 27,97011 0,873914 155,2484 561 30,82139 30,066 0,885644 165,8603 FDIinflow / GDP 1400 0,014534 0,091421 -0,25039 2,905041 575 0,010744 0,0165723 -0,00585 0,174687 PE inflow / GDP 1400 0,001116 0,0133449 -0,02336 0,393356 575 0,000623 0,0026516 -0,02027 0,030494 PD inflow / GDP 1400 0,002147 0,0072285 -0,05949 0,080272 575 0,001451 0,0060928 -0,19338 0,080271 OI inflow / GDP 1400 0,004028 0,0282838 -0,51155 0,187963 575 0,003637 0,0161132 -0,07653 0,147741 FDI outflow / GDP 1400 0,005736 0,0893958 -0,30511 2,952914 575 0,001555 0,0064208 -0,01834 0,080312 PE outflow / GDP 1400 0,001492 0,0240897 -0,27779 0,825351 575 0,000277 0,0012912 -0,007 0,013604 PD outflow / GDP 1400 0,001259 0,0061015 -0,44037 0,092667 575 0,000652 0,0024463 -0,0083 0,021443 OI outflow / GDP 1400 0,004817 0,0182729 -0,15411 0,341747 575 0,005006 0,0145559 -0,33515 0,143434

Capital account openness 1400 0,027565 0,1917131 -0,47103 5,317931 575 0,019122 0,0368262 -0,07336 0,377181

consumption (share of GDP) 1400 0,768437 0,2190494 0,197711 1,658968 575 0,713457 0,1883758 0,116647 1,404291

inflows x corruption 1344 68,14045 424,7658 -2810,1 11677,44 551 67,1683 137,6582 -441,929 1040,043

inflows x financial developemnt 1345 87,63668 801,1463 -1658,25 23454,15 561 46,44183 116,2294 -571,815 1036,507

inflows x QoG 1225 0,546867 1,78016 -15,6403 28,65714 508 0,544549 1,107506 -3,87704 7,90004

21 We will now dig a little deeper in to the variables of interest, the gross capital flows. Figure 3 and 4 show for the investment-constrained and savings-constrained countries total gross inflows and gross outflows in millions of US dollars. In the analysis we look at flows over GDP, but these graphs clearly show that over the past decades gross capital flows have increased rapidly. In the beginning of our sample, both capital inflows and outflows were nihil, but their value has increased rapidly over the past decades. Another observations that can be made here is that inflows into these developing countries are almost always met with a similar outflows by domestic agents. This is in line with the literature on the Lucas paradox and goes against the neoclassical theory.

Figure 3: Gross inflows by foreigners and gross outflows by domestic agents in investment-constrained economies

Figure 4: Gross inflows by foreigners and gross outflows by domestic agents in savings-constrained economies Figure 5 and 6 show the decomposition of inflows as percentages of GDP. These graphs show a similar story for both the investment-constrained and savings-constrained sample of

countries. Foreign direct investment inflows make up the biggest share of inflows, followed 0,0 100.000,0 200.000,0 300.000,0 400.000,0 500.000,0 600.000,0 700.000,0 800.000,0 900.000,0 1.000.000,0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Milli o n s o f cu rre n t U S d o llars Years

Gross flows - savings-constrained

CIF COD 0,0 100.000,0 200.000,0 300.000,0 400.000,0 500.000,0 600.000,0 700.000,0 800.000,0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Milli o n s o f cu rre n tUS d o llars Years

Gross flows - investment-constrained

22 by other investment inflows, portfolio debt inflows and portfolio equity inflows. We can also see some evidence that with the financial crisis, the inflows to investment constrained

economies seem to dry up more compared to the inflows into savings-constrained economies. Especially the line for other investments shows a massive drop, this is in line with the idea of the global financial cycle / global liquidity.

Figure 5: Decomposition of flows for investment-constrained economies

Figure 6: Decomposition of flows for savings-constrained economies

The last part of the descriptive statistics looks at the correlations between the variables used. The correlation table can be found in Appendix C. There are no big surprises within this table, there are a few variables that are highly correlated but those make sense. For example, quality

-1,00% -0,50% 0,00% 0,50% 1,00% 1,50% 2,00% 2,50% 3,00% 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 G ro ss f lows % o f G DP Years

Decomposition of flows (% of GDP) - investment-constrained

Foreign direct investment Portfolio equity Portfolio debt Other investment

-0,50% 0,00% 0,50% 1,00% 1,50% 2,00% 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 % o f G DP Years

Decomposition flows - savings-constrained (% of GDP)

23 of government variable is correlated with the corruption variable. This makes sense since the corruption variable is a subset of quality of government variable. Furthermore, some of the different flow variables are also highly correlated with each other and especially with the interaction variables between total inflows and the threshold effects. This makes sense as well as those different flow variables make up total inflows.

Another problem could be multicollinearity, where independent variables are strongly

correlated with each other, due to which a variable can be linearly predicted by a combination of other variables. This could lead to higher standard errors and lower T statistics. To test for this we use the variance inflation factor (VIF), this postestimation cannot be used in a panel data model. Therefore, we run a OLS regression, after which we test for multicollinearity. We include all our variables in this regression (also the interaction with the savings-constrained dummy), the mean VIF is 3.36. However, there does seem to be a problem of

multicollinearity with the variable for capital account openness, and the three interactions with the thresholds(VIF ± 15). When we run the regressions again without these variables, the mean VIF is 1.80 and the highest individual score is 3.82 which indicates that

24 5. Data analysis

Now we come to the empirical part of this paper. Our sample include 79 countries, from which we subtract Mauritius due to the many outliers that country has. Furthermore, in most regressions some countries fall out due to missing data on some variables, that is the reason that the number of Id is most of the time close to but lower than the 78 countries.

From the methodology, we know that there is no absolute consensus on two parts of the model. The first is whether to use the control variables in levels or in differences. Secondly, how many years/periods should the lags for the capital flows be.

Table 3, in Appendix D shows the regressions results for the annual fixed effects model. Column 1 shows the baseline model, only the control variables, where these control variables are taken in differences. The second column does the same analysis, but now the controls are in levels. From these results, we can see that the model in column 2 is a better fit, this means that we will in further analyses use levels instead of differences. This was also the preferred specification as mentioned in the methodology section. Something that stands out from column 2 is that the variable for institutional quality is not even close to significance. The theoretical argument for this variables was not very strong to begin with, and therefore we will also not use this variable again in further analyses.

The second issue, regarding the lag structure of capital flows is not shown in the table. But we tried the model with different lags, two and three years, and found that with two years only foreign direct investment inflow was significant and with three year lags none of the flows were significant anymore. Therefore, in the rest of the analyses we will use a one year lag, or a one period lag when we are not using annual observations.

In column 3, we add the capital inflows and the interaction between inflows and the savings-constrained dummy. Here investment is not included, if investment is included the

significance of especially foreign direct investment inflows drops dramatically. This is due to the fact that foreign direct investment affects economic growth mainly through investment. Portfolio debt flows and other investment flows (debt flows) enter insignificantly and the interactions are even negative. Foreign direct investment inflow is not significant, but the interaction is. This means that these flows have a positive effect on economic growth in the savings-constrained countries. Portfolio equity flows do enter significantly, the interaction is not significant but does have a substantial coefficient. In column four, the different types of capital outflows are also added, some of which are positive others negative and mostly weakly significant. Column 5 and 6 show the result from the regressions separately for investment and savings-constrained economies. These two columns allow us to clearly compare the results for both groups. The first thing that stands out is that, the coefficient on the lagged dependent is a lot higher for the savings-constrained economies. This indicates that these countries grow faster, something that we also saw in the descriptive statistics. When we look at the coefficient on the different capital inflows, we see a similar picture as with the

25 The interpretations of the coefficients in this fixed-effects model depend on the variable in question. For example the coefficient of foreign direct investment inflows in column four (0.141) indicates that for investment-constrained economies, a one percent gross foreign direct investment inflows over GDP in the last year would lead to an 0.141% increase in economic growth (although this result was not significant). For savings-constrained, we then have to take both that coefficient and the one from the interaction term. This means that an one percent gross foreign direct investment inflows over GDP in the last year leads to a 0.578% increase in economic growth. The coefficients for portfolio equity inflows are quite large, but in the descriptive statistics in became clear that as a share of GDP these flows are not that big. Another way of interpreting these coefficients, which takes this into account, is by looking at the effect of one standard deviation increase in portfolio equity inflows. For example the coefficient for savings-constrained economies is 1.506 (column 6), from table 2 we then take the standard deviation for that variable, 0.26516 (we multiplied it by 100 to give the result in %). A one standard deviation increase in portfolio equity inflows over GDP then leads to 0.379% increase in economic growth in the next period. This is lower than the coefficient in the table, but is still a substantial impact.

A last observation that we can make on the basis of this table is that it seems that model is a better fit for the savings-constrained economies than for the investment-constrained

economies group. The R-squared adjusted is 0.313 compared to 0.194 in columns 6 and 5. The dataset allows us to split up the sample in many different ways, in table 4 and 5 in

Appendix D we show the results of two such ways. In the first we split up the sample in five 5 year periods. The analysis is then still an annual fixed effect model, but the period in each analysis is only five years. The results show that the effect of capital inflows, and the fit of the model, vary over the periods studied. However, the basic findings from above still seem to hold. Foreign direct investment and portfolio equity inflows show the most positive results, with the coefficient being higher for savings-constrained economies. In table 5, the sample is split into four different regions: Africa, Asia, America and Europe. Two countries are not included in any of these groups due to the ambiguity in which groups they should be, these are Azerbaijan and Jordan. In this analysis, only capital inflows are included. Some of the result here are surprising, for America and Europe the interaction terms between foreign direct investment inflows and the savings-constrained dummy are negative but not significant. Another surprise is that the coefficients on both foreign direct investment inflows and

portfolio equity inflows are higher for the African countries in our sample than for the Asian one. This is a surprise because the Asian countries are seen as very profitable places to invest in.

26 5.2 Robustness checks

The results above were all from fixed effects models with annual observations. The next step to check the robustness of these results is by averaging the observations over multiple years. This, as explained in the methodology, smooths out the effects of e.g. business cycles. Here both a fixed-effects model and the Arellano-Bond estimation (GMM) will be used. In table 6, we show the results for the fixed effects models with 4 and 6 year averaged periods.

The results for the different capital outflows are not shown, to limit the table in size. In the first 3 columns the averaged periods are 4 non-overlapping years and all flows are lagged one period. The results of the interaction model are in column 1, and the separate analysis for investment-constrained and savings-constrained economies are in columns 2 and 3. A couple a results stand out, the first being that the results for the capital flows are less robust

(significant) than for the analyses with annual observations. The coefficients on foreign direct investment inflows and portfolio equity inflows for investment-constrained economies are lower and sometimes significantly negative. Whereas for savings-constrained economies both still seem to have a positive, and substantial, effect on economic growth. Although the

coefficients are not always significant.

With the averaging of observations the interpretation of the coefficients also changes. Again the coefficient for foreign direct investment inflows from the interaction models can be taken as an example, although these coefficients are not significant. For investment-constrained economies, an average foreign direct investment inflow of one percent of GDP over the last period (4 years) leads to an increase in economic growth of 0.165% in this period. Again, for savings-constrained economies this is 0.165 + 0.596 = 0.761%.

However, the averaging of observations also leads to a problem, this became clear especially for the 6 year averaged periods (not shown in the table). Where all flows were insignificant. This could be due to the fact that the same weight is given to observations from the different years. For example, for the first period (1991-1996) capital flows are averaged and used to explain economic growth in the next period (1997-2002). Flows in 1991 now have the same weight as flows in 1996, whilst these are closer to and can more reasonably be expected to have an effect on economic growth in that period. Therefore, in column 4 the flows are not lagged one period but one year before being averaged. This means that for the period 1997-2002 the average flow over the period 1996-2001 is used. These results are again very similar to those of annual fixed effects model. With positive effects of foreign direct investment flows and portfolio equity flows, and higher coefficients for savings-constrained economies.

The last robustness check is using the Arellano Bond estimation (GMM) instead of the fixed-effects model. This estimation is getting more and more popular in the literature as it helps to solve the endogeneity problems that often arise with economic questions. This is possible by using instrumental variables, the number of instrumental variables can be seen in the bottom of table 7. The number of instruments can be very large, because the Arellano-Bond

27 Another difference in the table is the inclusion of the Arellano-Bond test, which is a test for zero autocorrelation in first-differenced errors. If this test shows a value < 0.05, then there is still a problem with endogeneity and the model should be specified differently. In table 7, we show the results for annual observations, 4 year averages and 6 year averages. Again we do not show the results for outflows, to make the table more legible. The outflow variable were also not consistent nor very significant. The interpretation of the coefficients does not change compared to the fixed-effects model.

Column 1 shows the interaction model for annual observations, the coefficient on foreign direct investment inflows is negative which is unexpected. Next to the coefficient for portfolio equity inflows, the coefficient on portfolio debt inflows is positive. The interaction with the savings-constrained dummy are positive for both FDI and portfolio equity inflows, where the one on FDI is also significant.

Columns 2 till 4 show the results for the 4 year averaged periods. Where the lags for the capital flows are one period. The interaction model in column 2 shows that for investment-constrained economies the coefficient on foreign direct investment inflows is close to zero, and the coefficient for portfolio equity flows is negative and significant. Whereas the coefficient on the interaction with the dummy are positive though not significant. The same results are also found in column 3 and 4, which show the results of the regressions for both groups separately. The last column show the result for the 6 year averaged period interaction model, again we do not lag the flows one period but one year before averaging the

observations. The coefficients on foreign direct investment and portfolio equity inflows are positive. Whereas, those for portfolio debt and other investment inflows are negative. The interactions with the savings-constrained dummy are all significant and substantial. With the same sign as the coefficients with interactions. This would indicate that equity like flows are more positive in savings-constrained economies and debt like flows more negative.

28 6. Conclusion / discussion

After the analysis we now come back to the research question of this paper, What is the effect

of capital flows on economic growth in savings versus investment constrained developing countries? This question, and linked to it the hypothesis, that capital inflows have a bigger

positive impact on savings-constrained economies compared to investment-constrained ones can now be answered. We will do so in this section, and compare the result to the results we found in the literature. After this some points will be discussed concerning the limitations, implications, and possible future research on this topic.

To answer the research question we used a sample of 79 developing countries, over the period 1990 – 2014. Our basic model to test the relation from the research question was an annual fixed-effects model. To test the robustness of these result other analyses were performed, such as fixed effects models for 4 and 6 year averaged periods and a generalized methods of

moments estimation (Arellano-Bond)

When we look at the hypothesis that capital inflow are more beneficial for

savings-constrained economies, we can say that we found some evidence in favour but that we cannot completely accept it. First of all, in the descriptive statistics we saw that savings-constrained economies do not have bigger shares of capital inflows to GDP, although we did see that it looked more stable compared to investment-constrained economies. Secondly, when we compare the coefficients on the different capital inflows throughout our analysis we can conclude that the effect on economic growth of portfolio debt inflows and other investment inflows are not greater for savings-constrained economies. However, we did find evidence that this is the case for foreign direct investment and portfolio equity inflows, the two types of flows that seem to be most beneficial. While they show positive results for both groups, the effect for savings-constrained economies seems larger. This is line with the surveyed literature, studies that look specifically at different capital flows often find that equity flows are good for economic growth whilst debt flows aren’t. However, some results were

unexpected and not in line with the literature. For example a few coefficients regarding the interaction with debt inflows and the savings-constrained dummy were significantly negative. This indicates that in these analysis debt inflows are more harmful in savings-constrained than in investment-constrained economies.

29 What is the implication of these results for policy in developing countries. The results

underline that the overly positive view on capital account liberalization based on neoclassical economics is too simple. The need to distinguish between different types of flows is probably the most important implications from this study, and there is also some support for the view that different developing countries are held back by different constraints. This means for policy that just opening up an country is no panacea, but the case for partially opening up to equity type flows (equity account liberalization) is quite strong.

This study, as most studies do, also has limitations. These limitations have to do with the sample of countries that we use and the limitations on data. The countries that are used in the sample are developing countries, but there are also many developing countries that are not included in our sample. It is likely that the countries that are missing are of the poorest variety with the most potential benefit from maybe opening up. It is hard to say if the conclusions from this study are also valid for those countries. Secondly, there are big limitations with regard to the data. This has several reasons, the first has to do with capital flow data, the distinction between the different types of flows has not always been the same which might make it harder to test certain relations. For example, over the past decades some flows that were earlier characterized as foreign direct investment flows are now counted as portfolio equity flows and vice versa. Then there is the big problem that for the developing countries for which we have data, the quality of the data is not always very good. This can mean that there are gaps in coverage for variables or that variables are under or overstated. Furthermore, for example the measurement of GDP depends on the models that the statistical agency uses, if these are updated then sometimes there is a sudden very big increase in GDP. This for example happened to quite a few African countries.

As mentioned in the policy implications, our results are only for developing countries that have a completely or partially open capital account. Therefore, for policymakers that have to decide if opening up is a good idea such research might be of limited use. Future research could therefore focus on the characteristics that savings-constrained and

investment-constrained countries have and that could be identified beforehand. Our results also suggest that research on equity account liberalization might be fruitful, especially if focused solely on developing countries. Now most research use samples of countries that include very few or no developing countries.

Another avenue could be looking into different ways to measure savings-constraints. In this study we use a dummy and a hard cut-off point. Creating a continuous variable, an index for example, could help make a distinction between mildly and very savings-constrained